Modifier and Type | Field and Description |
---|---|
cudnnHandle |
cudnnDebug.handle
cudnn handle
|
Modifier and Type | Method and Description |
---|---|
static int |
JCudnn.cudnnActivationBackward(cudnnHandle handle,
cudnnActivationDescriptor activationDesc,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
Function to perform backward activation
|
static int |
JCudnn.cudnnActivationForward(cudnnHandle handle,
cudnnActivationDescriptor activationDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform forward activation
|
static int |
JCudnn.cudnnAddTensor(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor aDesc,
Pointer A,
Pointer beta,
cudnnTensorDescriptor cDesc,
Pointer C)
Tensor Bias addition : C = alpha * A + beta * C
|
static int |
JCudnn.cudnnBackendExecute(cudnnHandle handle,
cudnnBackendDescriptor executionPlan,
cudnnBackendDescriptor variantPack) |
static int |
JCudnn.cudnnBatchNormalizationBackward(cudnnHandle handle,
int mode,
Pointer alphaDataDiff,
Pointer betaDataDiff,
Pointer alphaParamDiff,
Pointer betaParamDiff,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor dxDesc,
Pointer dx,
cudnnTensorDescriptor dBnScaleBiasDesc,
Pointer bnScale,
Pointer dBnScaleResult,
Pointer dBnBiasResult,
double epsilon,
Pointer savedMean,
Pointer savedInvVariance)
Performs backward pass of Batch Normalization layer.
|
static int |
JCudnn.cudnnBatchNormalizationBackwardEx(cudnnHandle handle,
int mode,
int bnOps,
Pointer alphaDataDiff,
Pointer betaDataDiff,
Pointer alphaParamDiff,
Pointer betaParamDiff,
cudnnTensorDescriptor xDesc,
Pointer xData,
cudnnTensorDescriptor yDesc,
Pointer yData,
cudnnTensorDescriptor dyDesc,
Pointer dyData,
cudnnTensorDescriptor dzDesc,
Pointer dzData,
cudnnTensorDescriptor dxDesc,
Pointer dxData,
cudnnTensorDescriptor dBnScaleBiasDesc,
Pointer bnScaleData,
Pointer bnBiasData,
Pointer dBnScaleData,
Pointer dBnBiasData,
double epsilon,
Pointer savedMean,
Pointer savedInvVariance,
cudnnActivationDescriptor activationDesc,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
JCudnn.cudnnBatchNormalizationForwardInference(cudnnHandle handle,
int mode,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor bnScaleBiasMeanVarDesc,
Pointer bnScale,
Pointer bnBias,
Pointer estimatedMean,
Pointer estimatedVariance,
double epsilon)
Performs Batch Normalization during Inference:
y[i] = bnScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + bnBias[k]
with bnScale, bnBias, runningMean, runningInvVariance tensors indexed
according to spatial or per-activation mode.
|
static int |
JCudnn.cudnnBatchNormalizationForwardTraining(cudnnHandle handle,
int mode,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor bnScaleBiasMeanVarDesc,
Pointer bnScale,
Pointer bnBias,
double exponentialAverageFactor,
Pointer resultRunningMean,
Pointer resultRunningVariance,
double epsilon,
Pointer resultSaveMean,
Pointer resultSaveInvVariance)
Computes y = BN(x).
|
static int |
JCudnn.cudnnBatchNormalizationForwardTrainingEx(cudnnHandle handle,
int mode,
int bnOps,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer xData,
cudnnTensorDescriptor zDesc,
Pointer zData,
cudnnTensorDescriptor yDesc,
Pointer yData,
cudnnTensorDescriptor bnScaleBiasMeanVarDesc,
Pointer bnScale,
Pointer bnBias,
double exponentialAverageFactor,
Pointer resultRunningMean,
Pointer resultRunningVariance,
double epsilon,
Pointer resultSaveMean,
Pointer resultSaveInvVariance,
cudnnActivationDescriptor activationDesc,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Computes y = relu(BN(x) + z).
|
static int |
JCudnn.cudnnBuildRNNDynamic(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int miniBatch) |
static int |
JCudnn.cudnnConvolutionBackwardBias(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor dyDesc,
Pointer dy,
Pointer beta,
cudnnTensorDescriptor dbDesc,
Pointer db)
Function to compute the bias gradient for batch convolution
|
static int |
JCudnn.cudnnConvolutionBackwardData(cudnnHandle handle,
Pointer alpha,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx) |
static int |
JCudnn.cudnnConvolutionBackwardFilter(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer beta,
cudnnFilterDescriptor dwDesc,
Pointer dw) |
static int |
JCudnn.cudnnConvolutionBiasActivationForward(cudnnHandle handle,
Pointer alpha1,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer alpha2,
cudnnTensorDescriptor zDesc,
Pointer z,
cudnnTensorDescriptor biasDesc,
Pointer bias,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor yDesc,
Pointer y)
Fused conv/bias/activation operation : y = Act( alpha1 * conv(x) + alpha2 * z + bias )
|
static int |
JCudnn.cudnnConvolutionForward(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform the forward pass for batch convolution
|
static int |
JCudnn.cudnnCreate(cudnnHandle handle) |
static int |
JCudnn.cudnnCTCLoss_v8(cudnnHandle handle,
int algo,
cudnnCTCLossDescriptor ctcLossDesc,
cudnnTensorDescriptor probsDesc,
Pointer probs,
Pointer labels,
Pointer labelLengths,
Pointer inputLengths,
Pointer costs,
cudnnTensorDescriptor gradientsDesc,
Pointer gradients,
long workSpaceSizeInBytes,
Pointer workspace)
return the ctc costs and gradients, given the probabilities and labels
|
static int |
JCudnn.cudnnCTCLoss(cudnnHandle handle,
cudnnTensorDescriptor probsDesc,
Pointer probs,
int[] hostLabels,
int[] hostLabelLengths,
int[] hostInputLengths,
Pointer costs,
cudnnTensorDescriptor gradientsDesc,
Pointer gradients,
int algo,
cudnnCTCLossDescriptor ctcLossDesc,
Pointer workspace,
long workSpaceSizeInBytes)
return the ctc costs and gradients, given the probabilities and labels
|
static int |
JCudnn.cudnnDestroy(cudnnHandle handle) |
static int |
JCudnn.cudnnDivisiveNormalizationBackward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int mode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer means,
Pointer dy,
Pointer temp,
Pointer temp2,
Pointer beta,
cudnnTensorDescriptor dXdMeansDesc,
Pointer dx,
Pointer dMeans) |
static int |
JCudnn.cudnnDivisiveNormalizationForward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int mode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer means,
Pointer temp,
Pointer temp2,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
LCN/divisive normalization functions: y = alpha * normalize(x) + beta * y
|
static int |
JCudnn.cudnnDropoutBackward(cudnnHandle handle,
cudnnDropoutDescriptor dropoutDesc,
cudnnTensorDescriptor dydesc,
Pointer dy,
cudnnTensorDescriptor dxdesc,
Pointer dx,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
JCudnn.cudnnDropoutForward(cudnnHandle handle,
cudnnDropoutDescriptor dropoutDesc,
cudnnTensorDescriptor xdesc,
Pointer x,
cudnnTensorDescriptor ydesc,
Pointer y,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
JCudnn.cudnnDropoutGetStatesSize(cudnnHandle handle,
long[] sizeInBytes)
helper function to determine size of the states to be passed to cudnnSetDropoutDescriptor
|
static int |
JCudnn.cudnnFindConvolutionBackwardDataAlgorithm(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf[] perfResults) |
static int |
JCudnn.cudnnFindConvolutionBackwardDataAlgorithmEx(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
Pointer dx,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf[] perfResults,
Pointer workSpace,
long workSpaceSizeInBytes) |
static int |
JCudnn.cudnnFindConvolutionBackwardFilterAlgorithm(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor dwDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf[] perfResults) |
static int |
JCudnn.cudnnFindConvolutionBackwardFilterAlgorithmEx(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor dyDesc,
Pointer y,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor dwDesc,
Pointer dw,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf[] perfResults,
Pointer workSpace,
long workSpaceSizeInBytes) |
static int |
JCudnn.cudnnFindConvolutionForwardAlgorithm(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf[] perfResults) |
static int |
JCudnn.cudnnFindConvolutionForwardAlgorithmEx(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
Pointer y,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf[] perfResults,
Pointer workSpace,
long workSpaceSizeInBytes) |
static int |
JCudnn.cudnnFindRNNBackwardDataAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor[] dyDesc,
Pointer dy,
cudnnTensorDescriptor dhyDesc,
Pointer dhy,
cudnnTensorDescriptor dcyDesc,
Pointer dcy,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnTensorDescriptor[] dxDesc,
Pointer dx,
cudnnTensorDescriptor dhxDesc,
Pointer dhx,
cudnnTensorDescriptor dcxDesc,
Pointer dcx,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnFindRNNBackwardWeightsAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor[] yDesc,
Pointer y,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes,
cudnnFilterDescriptor dwDesc,
Pointer dw,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnFindRNNForwardInferenceAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnFindRNNForwardTrainingAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnFusedOpsExecute(cudnnHandle handle,
cudnnFusedOpsPlan plan,
cudnnFusedOpsVariantParamPack varPack) |
static int |
JCudnn.cudnnGetAlgorithmSpaceSize(cudnnHandle handle,
cudnnAlgorithmDescriptor algoDesc,
long[] algoSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnGetBatchNormalizationBackwardExWorkspaceSize(cudnnHandle handle,
int mode,
int bnOps,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor yDesc,
cudnnTensorDescriptor dyDesc,
cudnnTensorDescriptor dzDesc,
cudnnTensorDescriptor dxDesc,
cudnnTensorDescriptor dBnScaleBiasDesc,
cudnnActivationDescriptor activationDesc,
long[] sizeInBytes) |
static int |
JCudnn.cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize(cudnnHandle handle,
int mode,
int bnOps,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor zDesc,
cudnnTensorDescriptor yDesc,
cudnnTensorDescriptor bnScaleBiasMeanVarDesc,
cudnnActivationDescriptor activationDesc,
long[] sizeInBytes)
output means differential, can be NULL
|
static int |
JCudnn.cudnnGetBatchNormalizationTrainingExReserveSpaceSize(cudnnHandle handle,
int mode,
int bnOps,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor xDesc,
long[] sizeInBytes) |
static int |
JCudnn.cudnnGetConvolutionBackwardDataAlgorithm_v7(cudnnHandle handle,
cudnnFilterDescriptor filterDesc,
cudnnTensorDescriptor diffDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor gradDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf[] perfResults) |
static int |
JCudnn.cudnnGetConvolutionBackwardDataAlgorithmMaxCount(cudnnHandle handle,
int[] count) |
static int |
JCudnn.cudnnGetConvolutionBackwardDataWorkspaceSize(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
int algo,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the convolution given an algo
|
static int |
JCudnn.cudnnGetConvolutionBackwardFilterAlgorithm_v7(cudnnHandle handle,
cudnnTensorDescriptor srcDesc,
cudnnTensorDescriptor diffDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor gradDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf[] perfResults) |
static int |
JCudnn.cudnnGetConvolutionBackwardFilterAlgorithmMaxCount(cudnnHandle handle,
int[] count) |
static int |
JCudnn.cudnnGetConvolutionBackwardFilterWorkspaceSize(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor gradDesc,
int algo,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the convolution given an algo
|
static int |
JCudnn.cudnnGetConvolutionForwardAlgorithm_v7(cudnnHandle handle,
cudnnTensorDescriptor srcDesc,
cudnnFilterDescriptor filterDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor destDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf[] perfResults) |
static int |
JCudnn.cudnnGetConvolutionForwardAlgorithmMaxCount(cudnnHandle handle,
int[] count)
helper function to provide the convolution forward algo that fit best the requirement
|
static int |
JCudnn.cudnnGetConvolutionForwardWorkspaceSize(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
int algo,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the convolution given an algo
|
static int |
JCudnn.cudnnGetCTCLossWorkspaceSize_v8(cudnnHandle handle,
int algo,
cudnnCTCLossDescriptor ctcLossDesc,
cudnnTensorDescriptor probsDesc,
cudnnTensorDescriptor gradientsDesc,
long[] sizeInBytes)
return the workspace size needed for ctc
|
static int |
JCudnn.cudnnGetCTCLossWorkspaceSize(cudnnHandle handle,
cudnnTensorDescriptor probsDesc,
cudnnTensorDescriptor gradientsDesc,
int[] labels,
int[] labelLengths,
int[] inputLengths,
int algo,
cudnnCTCLossDescriptor ctcLossDesc,
long[] sizeInBytes)
return the workspace size needed for ctc
|
static int |
JCudnn.cudnnGetDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc,
cudnnHandle handle,
float[] dropout,
Pointer states,
long[] seed) |
static int |
JCudnn.cudnnGetFoldedConvBackwardDataDescriptors(cudnnHandle handle,
cudnnFilterDescriptor filterDesc,
cudnnTensorDescriptor diffDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor gradDesc,
int transformFormat,
cudnnFilterDescriptor foldedFilterDesc,
cudnnTensorDescriptor paddedDiffDesc,
cudnnConvolutionDescriptor foldedConvDesc,
cudnnTensorDescriptor foldedGradDesc,
cudnnTensorTransformDescriptor filterFoldTransDesc,
cudnnTensorTransformDescriptor diffPadTransDesc,
cudnnTensorTransformDescriptor gradFoldTransDesc,
cudnnTensorTransformDescriptor gradUnfoldTransDesc)
Helper function to calculate folding descriptors for dgrad
|
static int |
JCudnn.cudnnGetMultiHeadAttnBuffers(cudnnHandle handle,
cudnnAttnDescriptor attnDesc,
long[] weightSizeInBytes,
long[] workSpaceSizeInBytes,
long[] reserveSpaceSizeInBytes) |
static int |
JCudnn.cudnnGetMultiHeadAttnWeights(cudnnHandle handle,
cudnnAttnDescriptor attnDesc,
int wKind,
long weightSizeInBytes,
Pointer weights,
cudnnTensorDescriptor wDesc,
Pointer wAddr) |
static int |
JCudnn.cudnnGetNormalizationBackwardWorkspaceSize(cudnnHandle handle,
int mode,
int normOps,
int algo,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor yDesc,
cudnnTensorDescriptor dyDesc,
cudnnTensorDescriptor dzDesc,
cudnnTensorDescriptor dxDesc,
cudnnTensorDescriptor dNormScaleBiasDesc,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor normMeanVarDesc,
long[] sizeInBytes,
int groupCnt)
Place hold for future work, should be set to 1 now
|
static int |
JCudnn.cudnnGetNormalizationForwardTrainingWorkspaceSize(cudnnHandle handle,
int mode,
int normOps,
int algo,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor zDesc,
cudnnTensorDescriptor yDesc,
cudnnTensorDescriptor normScaleBiasDesc,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor normMeanVarDesc,
long[] sizeInBytes,
int groupCnt) |
static int |
JCudnn.cudnnGetNormalizationTrainingReserveSpaceSize(cudnnHandle handle,
int mode,
int normOps,
int algo,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor xDesc,
long[] sizeInBytes,
int groupCnt)
Place hold for future work, should be set to 1 now
|
static int |
JCudnn.cudnnGetReductionIndicesSize(cudnnHandle handle,
cudnnReduceTensorDescriptor reduceTensorDesc,
cudnnTensorDescriptor aDesc,
cudnnTensorDescriptor cDesc,
long[] sizeInBytes)
Helper function to return the minimum size of the index space to be passed to the reduction given the input and
output tensors
|
static int |
JCudnn.cudnnGetReductionWorkspaceSize(cudnnHandle handle,
cudnnReduceTensorDescriptor reduceTensorDesc,
cudnnTensorDescriptor aDesc,
cudnnTensorDescriptor cDesc,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the reduction given the input and output
tensors
|
static int |
JCudnn.cudnnGetRNNBackwardDataAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNBackwardWeightsAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNDescriptor_v6(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] hiddenSize,
int[] numLayers,
cudnnDropoutDescriptor dropoutDesc,
int[] inputMode,
int[] direction,
int[] cellMode,
int[] algo,
int[] mathPrec)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNForwardInferenceAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNForwardTrainingAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNLinLayerBiasParams(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int pseudoLayer,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
Pointer w,
int linLayerID,
cudnnFilterDescriptor linLayerBiasDesc,
Pointer linLayerBias)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNLinLayerMatrixParams(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int pseudoLayer,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
Pointer w,
int linLayerID,
cudnnFilterDescriptor linLayerMatDesc,
Pointer linLayerMat)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNParamsSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnTensorDescriptor xDesc,
long[] sizeInBytes,
int dataType)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNProjectionLayers(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] recProjSize,
int[] outProjSize)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNTempSpaceSizes(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int fMode,
cudnnRNNDataDescriptor xDesc,
long[] workSpaceSize,
long[] reserveSpaceSize) |
static int |
JCudnn.cudnnGetRNNTrainingReserveSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
long[] sizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnGetRNNWeightParams(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int pseudoLayer,
long weightSpaceSize,
Pointer weightSpace,
int linLayerID,
cudnnTensorDescriptor mDesc,
Pointer mAddr,
cudnnTensorDescriptor bDesc,
Pointer bAddr) |
static int |
JCudnn.cudnnGetRNNWeightSpaceSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
long[] weightSpaceSize) |
static int |
JCudnn.cudnnGetRNNWorkspaceSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
long[] sizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnGetStream(cudnnHandle handle,
cudaStream_t streamId) |
static int |
JCudnn.cudnnIm2Col(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
Pointer colBuffer) |
static int |
JCudnn.cudnnLRNCrossChannelBackward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int lrnMode,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
LRN cross-channel backward computation.
|
static int |
JCudnn.cudnnLRNCrossChannelForward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int lrnMode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
LRN cross-channel forward computation.
|
static int |
JCudnn.cudnnMakeFusedOpsPlan(cudnnHandle handle,
cudnnFusedOpsPlan plan,
cudnnFusedOpsConstParamPack constPack,
long[] workspaceSizeInBytes) |
static int |
JCudnn.cudnnMultiHeadAttnBackwardData(cudnnHandle handle,
cudnnAttnDescriptor attnDesc,
int[] loWinIdx,
int[] hiWinIdx,
int[] devSeqLengthsDQDO,
int[] devSeqLengthsDKDV,
cudnnSeqDataDescriptor doDesc,
Pointer dout,
cudnnSeqDataDescriptor dqDesc,
Pointer dqueries,
Pointer queries,
cudnnSeqDataDescriptor dkDesc,
Pointer dkeys,
Pointer keys,
cudnnSeqDataDescriptor dvDesc,
Pointer dvalues,
Pointer values,
long weightSizeInBytes,
Pointer weights,
long workSpaceSizeInBytes,
Pointer workSpace,
long reserveSpaceSizeInBytes,
Pointer reserveSpace) |
static int |
JCudnn.cudnnMultiHeadAttnBackwardWeights(cudnnHandle handle,
cudnnAttnDescriptor attnDesc,
int addGrad,
cudnnSeqDataDescriptor qDesc,
Pointer queries,
cudnnSeqDataDescriptor kDesc,
Pointer keys,
cudnnSeqDataDescriptor vDesc,
Pointer values,
cudnnSeqDataDescriptor doDesc,
Pointer dout,
long weightSizeInBytes,
Pointer weights,
Pointer dweights,
long workSpaceSizeInBytes,
Pointer workSpace,
long reserveSpaceSizeInBytes,
Pointer reserveSpace) |
static int |
JCudnn.cudnnMultiHeadAttnForward(cudnnHandle handle,
cudnnAttnDescriptor attnDesc,
int currIdx,
int[] loWinIdx,
int[] hiWinIdx,
int[] devSeqLengthsQO,
int[] devSeqLengthsKV,
cudnnSeqDataDescriptor qDesc,
Pointer queries,
Pointer residuals,
cudnnSeqDataDescriptor kDesc,
Pointer keys,
cudnnSeqDataDescriptor vDesc,
Pointer values,
cudnnSeqDataDescriptor oDesc,
Pointer out,
long weightSizeInBytes,
Pointer weights,
long workSpaceSizeInBytes,
Pointer workSpace,
long reserveSpaceSizeInBytes,
Pointer reserveSpace) |
static int |
JCudnn.cudnnNormalizationBackward(cudnnHandle handle,
int mode,
int normOps,
int algo,
Pointer alphaDataDiff,
Pointer betaDataDiff,
Pointer alphaParamDiff,
Pointer betaParamDiff,
cudnnTensorDescriptor xDesc,
Pointer xData,
cudnnTensorDescriptor yDesc,
Pointer yData,
cudnnTensorDescriptor dyDesc,
Pointer dyData,
cudnnTensorDescriptor dzDesc,
Pointer dzData,
cudnnTensorDescriptor dxDesc,
Pointer dxData,
cudnnTensorDescriptor dNormScaleBiasDesc,
Pointer normScaleData,
Pointer normBiasData,
Pointer dNormScaleData,
Pointer dNormBiasData,
double epsilon,
cudnnTensorDescriptor normMeanVarDesc,
Pointer savedMean,
Pointer savedInvVariance,
cudnnActivationDescriptor activationDesc,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes,
int groupCnt)
Place hold for future work, should be set to 1 now
|
static int |
JCudnn.cudnnNormalizationForwardInference(cudnnHandle handle,
int mode,
int normOps,
int algo,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor normScaleBiasDesc,
Pointer normScale,
Pointer normBias,
cudnnTensorDescriptor normMeanVarDesc,
Pointer estimatedMean,
Pointer estimatedVariance,
cudnnTensorDescriptor zDesc,
Pointer z,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor yDesc,
Pointer y,
double epsilon,
int groupCnt)
Performs Normalization during Inference:
y[i] = normScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + normBias[k]
with normScale, normBias, runningMean, runningInvVariance tensors indexed
according to per-channel or per-activation mode.
|
static int |
JCudnn.cudnnNormalizationForwardTraining(cudnnHandle handle,
int mode,
int normOps,
int algo,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer xData,
cudnnTensorDescriptor normScaleBiasDesc,
Pointer normScale,
Pointer normBias,
double exponentialAverageFactor,
cudnnTensorDescriptor normMeanVarDesc,
Pointer resultRunningMean,
Pointer resultRunningVariance,
double epsilon,
Pointer resultSaveMean,
Pointer resultSaveInvVariance,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor zDesc,
Pointer zData,
cudnnTensorDescriptor yDesc,
Pointer yData,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes,
int groupCnt)
Computes y = relu(Norm(x) + z).
|
static int |
JCudnn.cudnnOpTensor(cudnnHandle handle,
cudnnOpTensorDescriptor opTensorDesc,
Pointer alpha1,
cudnnTensorDescriptor aDesc,
Pointer A,
Pointer alpha2,
cudnnTensorDescriptor bDesc,
Pointer B,
Pointer beta,
cudnnTensorDescriptor cDesc,
Pointer C)
B tensor is ignored for CUDNN_OP_TENSOR_SQRT, CUDNN_OP_TENSOR_NOT.
|
static int |
JCudnn.cudnnPoolingBackward(cudnnHandle handle,
cudnnPoolingDescriptor poolingDesc,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
Function to perform backward pooling
|
static int |
JCudnn.cudnnPoolingForward(cudnnHandle handle,
cudnnPoolingDescriptor poolingDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform forward pooling
|
static int |
JCudnn.cudnnQueryRuntimeError(cudnnHandle handle,
int[] rstatus,
int mode,
cudnnRuntimeTag tag) |
static int |
JCudnn.cudnnReduceTensor(cudnnHandle handle,
cudnnReduceTensorDescriptor reduceTensorDesc,
Pointer indices,
long indicesSizeInBytes,
Pointer workspace,
long workspaceSizeInBytes,
Pointer alpha,
cudnnTensorDescriptor aDesc,
Pointer A,
Pointer beta,
cudnnTensorDescriptor cDesc,
Pointer C)
The indices space is ignored for reduce ops other than min or max.
|
static int |
JCudnn.cudnnReorderFilterAndBias(cudnnHandle handle,
cudnnFilterDescriptor filterDesc,
int reorderType,
Pointer filterData,
Pointer reorderedFilterData,
int reorderBias,
Pointer biasData,
Pointer reorderedBiasData) |
static int |
JCudnn.cudnnRestoreAlgorithm(cudnnHandle handle,
Pointer algoSpace,
long algoSpaceSizeInBytes,
cudnnAlgorithmDescriptor algoDesc)
Deprecated.
|
static int |
JCudnn.cudnnRestoreDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc,
cudnnHandle handle,
float dropout,
Pointer states,
long stateSizeInBytes,
long seed)
Restores the dropout descriptor to a previously saved-off state
|
static int |
JCudnn.cudnnRNNBackwardData_v8(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] devSeqLengths,
cudnnRNNDataDescriptor yDesc,
Pointer y,
Pointer dy,
cudnnRNNDataDescriptor xDesc,
Pointer dx,
cudnnTensorDescriptor hDesc,
Pointer hx,
Pointer dhy,
Pointer dhx,
cudnnTensorDescriptor cDesc,
Pointer cx,
Pointer dcy,
Pointer dcx,
long weightSpaceSize,
Pointer weightSpace,
long workSpaceSize,
Pointer workSpace,
long reserveSpaceSize,
Pointer reserveSpace) |
static int |
JCudnn.cudnnRNNBackwardData(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor[] dyDesc,
Pointer dy,
cudnnTensorDescriptor dhyDesc,
Pointer dhy,
cudnnTensorDescriptor dcyDesc,
Pointer dcy,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnTensorDescriptor[] dxDesc,
Pointer dx,
cudnnTensorDescriptor dhxDesc,
Pointer dhx,
cudnnTensorDescriptor dcxDesc,
Pointer dcx,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNBackwardDataEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnRNNDataDescriptor yDesc,
Pointer y,
cudnnRNNDataDescriptor dyDesc,
Pointer dy,
cudnnRNNDataDescriptor dcDesc,
Pointer dcAttn,
cudnnTensorDescriptor dhyDesc,
Pointer dhy,
cudnnTensorDescriptor dcyDesc,
Pointer dcy,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnRNNDataDescriptor dxDesc,
Pointer dx,
cudnnTensorDescriptor dhxDesc,
Pointer dhx,
cudnnTensorDescriptor dcxDesc,
Pointer dcx,
cudnnRNNDataDescriptor dkDesc,
Pointer dkeys,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNBackwardWeights_v8(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int addGrad,
int[] devSeqLengths,
cudnnRNNDataDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor hDesc,
Pointer hx,
cudnnRNNDataDescriptor yDesc,
Pointer y,
long weightSpaceSize,
Pointer dweightSpace,
long workSpaceSize,
Pointer workSpace,
long reserveSpaceSize,
Pointer reserveSpace) |
static int |
JCudnn.cudnnRNNBackwardWeights(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor[] yDesc,
Pointer y,
Pointer workSpace,
long workSpaceSizeInBytes,
cudnnFilterDescriptor dwDesc,
Pointer dw,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNBackwardWeightsEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnRNNDataDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnRNNDataDescriptor yDesc,
Pointer y,
Pointer workSpace,
long workSpaceSizeInBytes,
cudnnFilterDescriptor dwDesc,
Pointer dw,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNForward(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int fwdMode,
int[] devSeqLengths,
cudnnRNNDataDescriptor xDesc,
Pointer x,
cudnnRNNDataDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor hDesc,
Pointer hx,
Pointer hy,
cudnnTensorDescriptor cDesc,
Pointer cx,
Pointer cy,
long weightSpaceSize,
Pointer weightSpace,
long workSpaceSize,
Pointer workSpace,
long reserveSpaceSize,
Pointer reserveSpace) |
static int |
JCudnn.cudnnRNNForwardInference(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
Pointer workSpace,
long workSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNForwardInferenceEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnRNNDataDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnRNNDataDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
cudnnRNNDataDescriptor kDesc,
Pointer keys,
cudnnRNNDataDescriptor cDesc,
Pointer cAttn,
cudnnRNNDataDescriptor iDesc,
Pointer iAttn,
cudnnRNNDataDescriptor qDesc,
Pointer queries,
Pointer workSpace,
long workSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNForwardTraining(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNForwardTrainingEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnRNNDataDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnRNNDataDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
cudnnRNNDataDescriptor kDesc,
Pointer keys,
cudnnRNNDataDescriptor cDesc,
Pointer cAttn,
cudnnRNNDataDescriptor iDesc,
Pointer iAttn,
cudnnRNNDataDescriptor qDesc,
Pointer queries,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnRNNGetClip(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] clipMode,
int[] clipNanOpt,
double[] lclip,
double[] rclip)
Deprecated.
|
static int |
JCudnn.cudnnRNNSetClip(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int clipMode,
int clipNanOpt,
double lclip,
double rclip)
Deprecated.
|
static int |
JCudnn.cudnnSaveAlgorithm(cudnnHandle handle,
cudnnAlgorithmDescriptor algoDesc,
Pointer algoSpace,
long algoSpaceSizeInBytes)
Deprecated.
|
static int |
JCudnn.cudnnScaleTensor(cudnnHandle handle,
cudnnTensorDescriptor yDesc,
Pointer y,
Pointer alpha)
Scale all values of a tensor by a given factor : y[i] = alpha * y[i]
|
static int |
JCudnn.cudnnSetDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc,
cudnnHandle handle,
float dropout,
Pointer states,
long stateSizeInBytes,
long seed) |
static int |
JCudnn.cudnnSetRNNAlgorithmDescriptor(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnAlgorithmDescriptor algoDesc)
Deprecated.
|
static int |
JCudnn.cudnnSetRNNDescriptor_v6(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int hiddenSize,
int numLayers,
cudnnDropoutDescriptor dropoutDesc,
int inputMode,
int direction,
int cellMode,
int algo,
int mathPrec)
Deprecated.
|
static int |
JCudnn.cudnnSetRNNProjectionLayers(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int recProjSize,
int outProjSize)
Deprecated.
|
static int |
JCudnn.cudnnSetStream(cudnnHandle handle,
cudaStream_t streamId) |
static int |
JCudnn.cudnnSetTensor(cudnnHandle handle,
cudnnTensorDescriptor yDesc,
Pointer y,
Pointer valuePtr)
Set all values of a tensor to a given value : y[i] = value[0]
|
static int |
JCudnn.cudnnSoftmaxBackward(cudnnHandle handle,
int algo,
int mode,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
Function to perform backward softmax
|
static int |
JCudnn.cudnnSoftmaxForward(cudnnHandle handle,
int algo,
int mode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform forward softmax
|
static int |
JCudnn.cudnnSpatialTfGridGeneratorBackward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer dgrid,
Pointer dtheta)
Place hold for future work, should be set to 1 now
|
static int |
JCudnn.cudnnSpatialTfGridGeneratorForward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer theta,
Pointer grid) |
static int |
JCudnn.cudnnSpatialTfSamplerBackward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx,
Pointer alphaDgrid,
cudnnTensorDescriptor dyDesc,
Pointer dy,
Pointer grid,
Pointer betaDgrid,
Pointer dgrid) |
static int |
JCudnn.cudnnSpatialTfSamplerForward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer grid,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y) |
static int |
JCudnn.cudnnTransformFilter(cudnnHandle handle,
cudnnTensorTransformDescriptor transDesc,
Pointer alpha,
cudnnFilterDescriptor srcDesc,
Pointer srcData,
Pointer beta,
cudnnFilterDescriptor destDesc,
Pointer destData) |
static int |
JCudnn.cudnnTransformTensor(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Tensor layout conversion helper (y = alpha * x + beta * y)
|
static int |
JCudnn.cudnnTransformTensorEx(cudnnHandle handle,
cudnnTensorTransformDescriptor transDesc,
Pointer alpha,
cudnnTensorDescriptor srcDesc,
Pointer srcData,
Pointer beta,
cudnnTensorDescriptor destDesc,
Pointer destData) |
Constructor and Description |
---|
cudnnDebug(int cudnn_version,
int cudnnStatus,
int time_sec,
int time_usec,
int time_delta,
cudnnHandle handle,
cudaStream_t stream,
long pid,
long tid,
int cudaDeviceId,
int[] reserved)
Creates a new cudnnDebug with the given values
|
Copyright © 2020. All rights reserved.